Accurate Power Forecasting Research Articles

Quantile-transformed multi-attention residual framework (QT-MARF) for medium-term PV and wind power prediction

The increasing generation of renewable electrical power, particularly from wind and solar sources, has significantly influenced the national energy and power transmission systems. However, accurate forecasting of wind and photovoltaic (PV) power remains challenging due to the stochastic and highly nonlinear nature of wind speed and solar irradiance. Traditional models often fail to produce accurate power forecasts. To address this challenge, this paper proposes a novel deep learning model based on the Quantile-Transformed Multi-Attention Residual Framework (QT-MARF). The proposed model is built on a Transformer architecture with Residual Net and Multi-Head Attention. QT-MARF utilizes sequential processing through gated residual networks, enabling the model to learn complex patterns and make accurate power forecasts. The model utilizes PV and wind data from Natal, Santa Vitoria, and the Chinese State Grid (CSG). Case studies are conducted to validate the estimation performance of the hybrid models. The proposed QT-MARF demonstrates promising results in terms of accuracy and efficiency, outperforming traditional models in metrics such as Mean Absolute Error (MAE), correlation coefficient (CC), Root Mean Squared Error (RMSE), and R-squared (R2). Comparative analysis with state-of-the-art techniques such as the Inception-embedded attention-based memory fully-connected network (IAMFN) model, CNN-GRU, CNN-LSTM, and RNN highlights the superiority of the proposed model. These findings suggest that the proposed model offers a promising solution for the challenging task of wind and PV power forecasting.

A Weak-Coupling Flow-Power Forecasting Method for Small Hydropower Station Group

Due to the need for rural revitalization and renewable energy utilization, a large quantity of small hydropower stations is emerging, with weak-coupling flow-power features. However, a weak spatial coupling exists between the distribution of small hydropower station groups (SHSGs) and gaging stations since the small hydropower stations are usually located in remote areas lacking hydrographic facilities. That may cause weak or no coupling between the hydroregime and the power output of small hydropower plants in the target basin, thus hindering accurate power forecasting. To meet the need for short-term power generation prediction for SHSGs in intensive management areas, we propose a data-driven power-forecasting model which can mine the correlation information of weakly coupled basins while transferring hydrological knowledge to uncoupled basins. First, to make the task data domains before and after migration more similar, a similar watershed matching algorithm based on the nonlinear dimensionality reduction algorithm (Isomap) and the k -means++ algorithm is proposed; then a short-term interpretable runoff prediction model is pretrained, and features are extracted in the source basin using the temporal fusion transformer (TFT) network. After that, a heuristic ensemble fine-tuning model based on the k -fold cross-validation fine-tuning method and heuristic ensemble algorithm is proposed to transfer the public knowledge of the source basin to the uncoupled basin. Then, a TFT network is used to mine the weak-coupling relationship between the hydrological regime and the output power of an SHSG. Finally, the validity of the model is verified with an example from a European region. After considering the weakly coupled flow-power characteristics, the mean absolute percentage error (MAPE) of three SHSGs’ power prediction by the proposed method is on average 34.8% lower than that of the method without considering the hydrological information.

Dual stream network with attention mechanism for photovoltaic power forecasting

The operations of renewable power generation systems highly depend on precise Photovoltaic (PV) power forecasting, providing significant economic, and environmental advantages for energy efficient buildings and urban energy systems. However, precise PV power forecasting, particularly, solar power is more challenging due to solar energy intermittence, instability, and randomness. These challenges hinder the integration of PV into smart grids, where accurate power forecasting is a promising solution in this direction, providing effective planning and management services. Therefore, in this work, we introduce a dual-stream network for accurate PV forecasting. The proposed network parallelly learns spatial patterns using convolutional network and temporal representations via sequential learning algorithm. These features are then integrated together to form a single, yet representative feature vector used as an input to self-attention mechanism to further select the optimal features for PV power forecasting. To the best of our knowledge, the proposed dual stream network with advanced features selection mechanism is a pioneering approach for time series analysis, narrowed towards PV power forecasting. We derive our network after a series of experimentations involving solo and hybrid models, resulting in higher forecasting accuracy against state-of-the-art models.

Individual yield nowcasting for residential PV systems

Due to the inherent uncertainty in photovoltaic (PV) energy generation, an accurate power forecasting is essential to ensure a reliable operation of PV systems and a safe electric grid. Machine learning (ML) techniques have gained popularity on the development of this task due to its increased accuracy. Most literature, however, focuses only on less than 5 PV systems during training process, which does not ensure generalization to unseen systems. When in presence of a large feet, regional forecasts are the norm. Nevertheless, none of these approaches are usable when it comes to monitoring residential PV systems. In this work, we propose a single ML model that is able to predict the individual power of a large fleet of 1102 PV systems. XGBoost algorithm was selected as the most suitable algorithm for the task of PV yield nowcasting due to its performance and ease of use. This algorithm obtains Mean Absolute Error (MAE) of 0.877 kWh (considering an average system size of 4.44 kWp) and Mean Absolute Percentage Error (MAPE) of 23% for hourly data aggregated to daily values. XGBoost predictions for individual PV systems are on average two times better than currently used commercial software. We discuss the lack of a suitable loss function that can combine absolute and relative errors for residential PV yield forecasting. We also point out the lack of an adequate metric to compute the error made on the predictions and provide hints on developing a suitable one.

Offshore Wind Power Forecasting—A New Hyperparameter Optimisation Algorithm for Deep Learning Models

The main obstacle against the penetration of wind power into the power grid is its high variability in terms of wind speed fluctuations. Accurate power forecasting, while making maintenance more efficient, leads to the profit maximisation of power traders, whether for a wind turbine or a wind farm. Machine learning (ML) models are recognised as an accurate and fast method of wind power prediction, but their accuracy depends on the selection of the correct hyperparameters. The incorrect choice of hyperparameters will make it impossible to extract the maximum performance of the ML models, which is attributed to the weakness of the forecasting models. This paper uses a novel optimisation algorithm to tune the long short-term memory (LSTM) model for short-term wind power forecasting. The proposed method improves the power prediction accuracy and accelerates the optimisation process. Historical power data of an offshore wind turbine in Scotland is utilised to validate the proposed method and compare its outcome with regular ML models tuned by grid search. The results revealed the significant effect of the optimisation algorithm on the forecasting models’ performance, with improvements of the RMSE of 7.89, 5.9, and 2.65 percent, compared to the persistence and conventional grid search-tuned Auto-Regressive Integrated Moving Average (ARIMA) and LSTM models.

Impact of the tilt angle, inverter sizing factor and row spacing on the photovoltaic power forecast accuracy

The growth of the installed photovoltaic (PV) capacity calls for accurate power forecasts, which are commonly calculated from irradiance forecasts using physical model chains. This study analyses the impact of the PV plant design parameters on the additional error of the irradiance-to-power conversion using three approaches. First, PV power curves are plotted for different design parameters to illustrate the amplification of errors and to enhance the understanding of the presented phenomena. Second, six forecast accuracy metrics are calculated as a function of the design parameters for the irradiance observation and forecast data of ten stations from five climate zones and two forecast providers. Third, the Pareto-optimal tradeoff between the annual energy production and the absolute forecast errors are obtained by the NSGA-II multi-objective optimization algorithm. The results reveal that the design parameters highly influence the PV power forecast errors, which is important to consider in all studies presenting PV power forecast verification. A tilt angle of 45° increases the error metrics up to 49% compared to a horizontal plane, and undersizing the inverters by a 1.5 sizing factor lead to an average error decrease up to 25%. Including the expected forecast errors as an objective during their design optimization process is important to maximize the net revenues of PV plants, especially if the unit imbalance costs are more than 50% of the electricity prices.

Intra-Day Solar Power Forecasting Strategy for Managing Virtual Power Plants

Solar energy penetration has been on the rise worldwide during the past decade, attracting a growing interest in solar power forecasting over short time horizons. The increasing integration of these resources without accurate power forecasts hinders the grid operation and discourages the use of this renewable resource. To overcome this problem, Virtual Power Plants (VPPs) provide a solution to centralize the management of several installations to minimize the forecasting error. This paper introduces a method to efficiently produce intra-day accurate Photovoltaic (PV) power forecasts at different locations, by using free and available information. Prediction intervals, which are based on the Mean Absolute Error (MAE), account for the forecast uncertainty which provides additional information about the VPP node power generation. The performance of the forecasting strategy has been verified against the power generated by a real PV installation, and a set of ground-based meteorological stations in geographical proximity have been used to emulate a VPP. The forecasting approach is based on a Long Short-Term Memory (LSTM) network and shows similar errors to those obtained with other deep learning methods published in the literature, offering a MAE performance of 44.19 W/m2 under different lead times and launch times. By applying this technique to 8 VPP nodes, the global error is reduced by 12.37% in terms of the MAE, showing huge potential in this environment.

Real-time power prediction approach for turbine using deep learning techniques

Accurate power forecasting is of great importance to the turbine control and predictive maintenance. However, traditional physics models and statistical models can no longer meet the needs of precision and flexibility when thermal power plants frequently undertake more and more peak and frequency modulation tasks. In this study, the recurrent neural network (RNN) and convolutional neural network (CNN) for power prediction are proposed, and are applied to predict real-time power of turbine based on DCS data (recorded for 719 days) from a power plant. In addition, the performances of two deep learning models and five typical machine learning models are compared, including prediction deviation, variance and time cost. It is found that deep learning models outperform other shallow models and RNN model performs best in balancing the accuracy-efficient trade-off for power prediction (the relative prediction error of 99.76% samples is less than 1% in all load range for test 216 days). Moreover, the influence of training size and input time-steps on the performance of RNN model is also explored. The model can achieve remarkable performance by learning only 30% samples (about 216 days) with 3 input time-steps (about 60 s). Those results of the proposed models based on deep-learning methods indicated that deep learning is of great help to improve the accuracy of turbine power prediction. It is therefore convinced that those models have a high potential for turbine control and predictable maintenance in actual industrial scenarios.

Advantages of digitalization in grid restoration

Abstract In Germany, the installed capacity of renewables exceeds the maximum load by far and the transmission task has changed from high-wind locations in the north to the economically strong south. In addition, the system inertia is reduced due to the decommissioning of large power stations. This leads to a higher probability of blackouts and a challenging restoration process. Operation during restoration differs strongly from normal. Besides other effects, the grid frequency is very sensitive to active power imbalances. An accurate power forecast of re-supplied feeders is required to estimate dynamic frequency deviations. Models for cold load pick-up (CLPU) and fluctuating renewable energy sources (RES) infeed are presented. For the coordination between Transmission system operators (TSO) and Distribution system operators (DSO) a clear and structured process is proposed to minimize the communication effort. Power system restoration (PSR) simulations are carried out and the results are presented.

Prediction Model for PV Performance With Correlation Analysis of Environmental Variables

With increasing installations of photovoltaic (PV) systems, interest in power forecasting has also increased. Inaccurate forecasts would result in substantial economic losses and system reliability issues. The correlation between weather variables and PV power is critical to ensure the efficient use of energy in PV systems. A key step toward accurate power forecasting is estimating the output from a PV system based on known environmental input data. In this research, all available weather data are used to predict the PV power. Meteorological and power data are then analyzed using a statistical approach to identify the order of significance of the input variables. Then, a predictive model is suggested as a function of irradiance, ambient temperature, wind speed, and relative humidity. The model produces a root mean square error of 4.957% and a mean absolute percentage error of 5.468% during the measurement period and over the entire range of irradiation.

A hybrid spatio-temporal forecasting of solar generating resources for grid integration

Recently, the installed solar generating resources have been increasing rapidly. Consequently, forecasting for solar generating resources are becoming an important work to integrate utility-scale solar generating resources into power systems. As solar generating resources are variable, uncontrollable, and uncertain, accurate and reliable forecasting enables higher penetrations of solar generating resources to be deployed on the electrical power grid. Accurate forecasting of solar resources contributes to evaluation of system reserves over large geographic area and to transmission system planning. To increase the penetration of solar generating resources on the electric power grid, the accurate power forecasting of geographically distributed solar generating resources is needed. In this paper, we propose a hybrid spatio-temporal forecasting of solar generating resources based on the naïve Bayesian classifier approach and spatial modelling approach. To validate our forecasting model, we use the empirical data from the practical solar farms in South Korea.

Microgrid energy and reserve management incorporating prosumer behind‐the‐meter resources

The unpredictable nature of renewable power, coupled with the influx of prosumers has made accurate power forecasts in microgrids (MGs) more difficult to obtain. A direct consequence of this is the need for additional spinning reserve (SR) capacity to compensate for resulting power imbalances. Due to the economic and environmental concerns, increasing conventional generation to meet this additional SR capacity is undesirable. The aggregation of prosumer behind-the-meter resources for the provision of SR is proposed in this study, and a mathematical model for the proposed scheme is developed. This scheme is formulated as a constrained optimisation problem, whose solution maintains power supply and demand balance whilst reserving a virtual spinning capacity. The formulation is linearised and solved using the CPLEX 12.6.3 solver in the Advanced Interactive Multidimensional Modelling System environment. A 14-bus MG test system is used to validate the proposed scheme, and results show the benefits of using prosumer behind-the-meter resources to provide ancillary services like SR.

Anisotropy stress invariants of thermally stratified wind turbine array boundary layers using large eddy simulations

In the interest of understanding the dynamics and energy transfer between the atmospheric boundary layer and large wind turbine arrays, a description of the turbulence anisotropy in the wake region of wind turbines is necessary. Flux of high momentum flow into the wind turbine array by anisotropic turbulence is a dominant factor of mean kinetic energy resupply for the wind farm. Under thermal stratification, the behavior of the turbulence field, and the energy flux, is significantly altered. Here, a thermally stratified wind turbine array boundary layer is studied in detail using a combination of Lumley and barycentric maps together with the recently introduced spheroid and color maps. The atmospheric flow is modelled using a large eddy simulation driven by a constant geostrophic wind and a time-varying ground surface temperature, obtained from a selected period of the Cooperative Atmosphere-Surface Exchange Study-99 field experiment. The wind farm is modelled using the traditional actuator-disk with rotation and yaw-alignment. The results show that turbulence under unstable stratification tends to be more isotropic than that under neutral and stable stratification. The turbulent mixing attributed to positive buoyancy in unstable regimes determines the energy distribution between flow layers, directly affecting the vertical distribution of anisotropy. Contrarily, in the stably stratified flow, negative buoyancy damps the turbulent fluctuations, hence affecting the evolution of the Reynolds stress, leading to an increase of flow anisotropy. In both thermal stratification regimes, wind turbines alter the structure of the turbulence within the atmospheric boundary layer by creating regions of greater anisotropy and expanding the boundary layer vertically. For the modeling community, it is important to note that the wind turbine wakes display a larger degree of anisotropy that is related well to the production of turbulent kinetic energy and mean kinetic energy entrainment, and hence, it is important to model correctly for accurate power forecasting. Leveraging the new insights into wind turbine/atmospheric boundary layer interaction found here will benefit the formulation of a new generation of efficient low order models for commercial application.

Wind Power Forecasting techniques in complex terrain: ANN vs. ANN-CFD hybrid approach

Due to technology developments, renewable energies are becoming competitive against fossil sources and the number of wind farms is growing, which have to be integrated into power grids. Therefore, accurate power forecast is needed and often operators are charged with penalties in case of imbalance. Yet, wind is a stochastic and very local phenomenon, and therefore hard to predict. It has a high variability in space and time and wind power forecast is challenging. Statistical methods, as Artificial Neural Networks (ANN), are often employed for power forecasting, but they have some shortcomings: they require data sets over several years and are not able to capture tails of wind power distributions. In this work a pure ANN power forecast is compared against a hybrid method, based on the combination of ANN and a physical method using computational fluid dynamics (CFD). The validation case is a wind farm sited in southern Italy in a very complex terrain, with a wide spread turbine layout.

A stochastic power curve for wind turbines with reduced variability using conditional copula

AbstractIt has been observed that a large variability exists between wind speed and wind power in real metrological conditions. To reduce this substantial variability, this study developed a stochastic wind turbine power curve by incorporating various exogenous factors. Four measurements, namely, wind azimuth, wind elevation, air density and solar radiation are chosen as exogenous influence factors. A recursive formula based on conditional copulas is used to capture the complex dependency structure between wind speed and wind power with reduced variability. A procedure of selecting a proper form for each factor and its corresponding copula models is given. Through a case study on the small wind turbine located in southeast of Edmonton, Alberta, Canada, we demonstrate that the variability can be reduced significantly by incorporating these influence factors. Wind turbine operators can apply the method reported in this study to construct a stochastic power curve for local wind farms and use it to achieve more accurate power forecasting and health condition monitoring of the turbine. Copyright © 2015 John Wiley & Sons, Ltd.